When an AI answer gets your founding date, headquarters or product description wrong, the fix is rarely in the AI product itself. It is in the public sources the answer was built from. Find the pages that state the wrong fact, make your own pages and profiles agree, and use the correction routes that platforms document. No public documentation we know of lets you edit a chatbot's answer directly.
This guide covers the sources worth checking, an order for fixing them, Organization markup for your own site, and where corrections can be requested.
Why the facts drift
Google describes its own system this way: Knowledge Graph facts come from a variety of sources that compile factual information, plus licensed data and information from content owners who suggest changes to claimed knowledge panels [1]. Knowledge panels are generated automatically, with information from various sources across the web [2].
The practical consequence: your company's facts exist in many places you did not write. If those places disagree, no single edit fixes the problem. A wrong date in an old directory listing and a correct date on your About page are both inputs.
We have no vendor documentation in hand that explains how each AI assistant chooses between conflicting sources, so this article does not claim to. The approach below does not depend on it. It removes the conflicts you can reach.
Step 1: Write down the canonical facts
Before checking anything, fix the facts in one place. A short internal record is enough.
| Fact | Canonical value | Notes |
|---|---|---|
| Legal name | Example Co, Inc. | Exact form |
| Brand name | Example Co | As used on the site |
| Founding date | 2016 | Year only, or full date, but pick one |
| Headquarters | 100 Main St, Austin, TX 78701, US | Same format everywhere |
| One-sentence description | "Example Co makes invoicing software for freelancers." | Reuse word for word |
| Official URL | https://www.example.com/ | One version, with or without www |
Decide ambiguous items now. If the company moved, or was incorporated in one year and launched in another, decide which date you call "founded" and say so wherever it appears.
Step 2: Find where the wrong fact lives
Search for the exact wrong value, not your company name. For a wrong founding year, search for "Example Co" "founded in 2014". Do the same for the wrong city or the wrong product description.
Then check each source type:
- Your own site. About page, footer, contact page, press page, older blog posts, and any
OrganizationJSON-LD. Old copy often survives in boilerplate "About us" paragraphs on press releases. - Google Business Profile and the knowledge panel, if you have either.
- LinkedIn company page.
- Wikipedia and Wikidata, if an entry exists for your company.
- Directories, app stores, marketplaces and social bios. Anything that carries a company description or address.
- Third-party articles. Older coverage that states the wrong fact. You usually cannot edit these, but you can note them.
Log each hit with its URL, the wrong value, and whether you control it.
Step 3: Make everything you control agree
Fix in this order, because the first items are the ones you fully own:
- About page and footer.
- Homepage, contact page and any press or "company" page.
- Organization structured data (below).
- LinkedIn. It asks for complete company information, including location and a primary office location, and super admins can edit the organization details [5].
- Google Business Profile, if you have one.
- Directories, app stores and social bios you can log in to.
Use the same name, address format and one-sentence description in all of them. Small variations such as "Inc." in one place and none in another, or "Austin" versus "Austin, TX", are harmless to a person but add noise when a system tries to match records.
Step 4: Add or correct Organization schema
Google recommends putting Organization structured data on the homepage or an about page, and using the same name and alternateName as the site name [3]. Its documentation lists properties such as address, logo, sameAs, telephone and numberOfEmployees, and says sameAs can point to social or review profiles [3]. Schema.org also defines foundingDate and description.
{
"@context": "https://schema.org",
"@type": "Organization",
"@id": "https://www.example.com/#organization",
"name": "Example Co",
"legalName": "Example Co, Inc.",
"url": "https://www.example.com/",
"logo": "https://www.example.com/logo.png",
"description": "Example Co makes invoicing software for freelancers.",
"foundingDate": "2016",
"address": {
"@type": "PostalAddress",
"streetAddress": "100 Main St",
"addressLocality": "Austin",
"addressRegion": "TX",
"postalCode": "78701",
"addressCountry": "US"
},
"sameAs": [
"https://www.linkedin.com/company/example-co"
]
}
Three rules for this block:
- Every value must match the visible text on the page. Markup that contradicts the page just adds another conflict.
- List only profiles you actually control or that are about your company in
sameAs. - Do not publish two Organization blocks with different names or dates on different pages.
For the markup test, see One Check Explained: Organization Schema. To tie Organization, WebSite and WebPage together with a shared @id, see JSON-LD @id: Connect Organization, WebSite, WebPage, Article. For validation steps and other types, see Schema.org Markup for AI Search.
Step 5: Request corrections where platforms allow them
Google Business Profile
Google says it gathers profile data from different sources, including user reports and licensed content, and may update profiles from those sources or from your public website [4]. Suggested edits may be published automatically if they are not reviewed within 4 days and are supported by other public information; some updates can be made directly by the owner [4].
Two actions follow. Fix your profile directly where you can. Then check the profile for suggested edits on a schedule, so a wrong one is not left to publish by default.
Knowledge panel
Google says some panel information comes from verified entities that have suggested edits to their own panels [2], and that content owners can suggest changes to claimed knowledge panels [1]. If a panel shows a wrong fact, claim and verify it, then suggest the edit. Keep the public sources from Step 3 consistent, because panels also draw on other sources across the web [2].
If you are a super admin, edit the organization details on the Page directly [5]. If you are not, ask whoever holds that role to do it, or have the current admin grant you access.
Wikipedia and Wikidata
First check whether your company has an entry in either. If not, there is nothing to correct there, and you should not create one just to fix a fact.
If an entry exists and is wrong:
- Wikidata items carry structured statements such as inception, headquarters location and official website. Compare each against your canonical record.
- Read the project's own guidance on editing about yourself or your employer, and disclose your connection as it requires.
- Where the guidance asks for it, propose the change on the article's talk page, or the equivalent discussion page, with a published source for the correct value.
Follow whichever process the project documents. Do not edit around it.
Directories and third-party sites
Most directories have a "claim this listing" or "report an error" route. Use it, and send the canonical record. For a news article with a wrong fact, a polite note to the publication asking for a correction is the normal route. Keep a list of what you sent and when.
Step 6: Recheck on a schedule
After the fixes, run the same questions again in the AI products your customers use, and run the exact-value searches from Step 2 again. Record the date, the question, and the answer.
No engine in our notes documents a timeline for when a corrected source reaches its answers, so pick a recheck schedule you can keep, such as monthly, and track whether the wrong value is shrinking across sources. A single answer is a snapshot. For a method that uses a dated change as a marker, see AI Search Freshness: How Fast a Changed Page Shows Up in Answers.
What to do next
- Write the canonical fact record: name, founding date, address, description, URL.
- Search for each wrong value and log every URL that carries it.
- Fix your site, Organization markup, LinkedIn and Google Business Profile first.
- Send correction requests for directories and any Wikipedia or Wikidata entry, following each platform's process.
- Set a recurring date to recheck both the sources and the AI answers.